The traditional interference suppression algorithms suffer from the shortcomings of relying on the accurate estimation of interference parameters in complex electromagnetic environments,needing the separate processing strategies for different types of interference,and weak generalization ability,This paper proposes a U-Net-based interference suppression algorithm using time-frequency image input.The proposed algorithm leverages the advantages of U-Net in image feature extraction and end-to-end learning by taking the short-time Fourier transform time-frequency image of the received signal as network input.The precise localization and effective suppression of interference components are achieved by using an encoding-decoding structure with skip connections,while preserving the original features of the target signal to the greatest extent.The proposed algorithm is compared with the conventional frequency-domain suppression methdod and the residual network-based suppression method in terms of four typical weak interferences,namely single-tone,multi-tone,narrowban and linear frequency modulation (LFM) interferences.The results show that the proposed U-Net-based algorithm exhibits an optimal bit error rate performance under different types of interference.It achieves approximately 1dB performance gain compared to traditional algorithms and about 0.5dB improvement over the residual network,and maintains stability under signal-to-interference ratios of 0dB and 5dB,thus verifying the strong robustness and good generalization of the proposed algorithm in weak interference environments.